Matrix equations of hydrodynamic type as lower-dimensional reductions of Self-dual type S-integrable systems A.I. Zenchuk 8 Center of Nonlinear Studies of L.D.Landau Institute for Theoretical Physics 0 0 (International Institute of Nonlinear Science) 2 n Kosygina 2, Moscow, Russia 119334 a E-mail: [email protected] J 1 1 February 1, 2008 ] I S . Abstract n i l n We show that matrix Q Q Self-dual type S-integrable Partial Differential Equations × [ (PDEs) possess a family of lower-dimensional reductions represented by the matrix Q × 3 n0Q quasilinear first order PDEs solved in [29] by the method of characteristics. In turn, v these PDEs admit two types of available particular solutions: (a) explicit solutions and 0 (b) solutions described implicitly by a system of non-differential equations. The later 5 0 solutions, in particular, exhibit the wave profile breaking. Only first type of solutions 2 is available for (1+1)-dimensional nonlinear S-integrable PDEs. (1+1)-dimensional N- . 8 wave equation, (2+1)- and (3+1)-dimensional Pohlmeyer equations are represented as 0 examples. We also represent a new version of the dressing method which supplies both 7 0 classical solutions and solutions with wave profile breaking to the above S-integrable : PDEs. v i X r 1 Introduction a There are several types of nonlinear equations of Mathematical Physics, which are referred to as integrable Partial Differential Equations (PDEs). We underline three following types of equations: 1. Equations integrable by the Inverse Spectral Transform Method (ISTM), which are also called soliton or S-integrable equations. Korteweg-de-Vries equation (KdV) has been discovered as the first representative of this class [1]. Among other methods of study of soliton equations we underline so-called dressing method, which was invented in [2, 3] and developed in set of papers, for instance, [4, 5, 6, 7, 8], see also [9, 10, 11]. 2. Nonlinear PDEs linearizable by some direct substitution, or C-integrable equations [12, 13, 14, 15, 16, 17]. Mostly remarkable is Hoph substitution allowing to integrate Bu¨rgers type nonlinear PDEs. 1 3. Quasilinear first order PDEs integrable by the method of characteristics [18] and by generalized hodograph method [19, 20, 21]. The later method is well applicable to (1+1)- dimensional nonlinear PDEs. Investigation of the higher dimensional systems by this method requires special efforts(see, forinstance, themethodof hydrodynamic reductions, which will be cited below). Many other method for study of the nonlinear PDEs have been developed. For instance, the algebraic-geometrical approach is well suited for the construction of local solutions [22, 23]. However, only restricted class of global solutions may be described in this way. A method for implicit description of solutions manifold is the method of hydrodynamic reductions [24, 25, 26, 27, 28], where solutions of multidimensional integrable PDEs are described in terms of Riemann invariants of some (1+1)-dimensional first order quasilinear PDEs. The criterion for applicability of such technique to the given nonlinear PDE has been invented. This method allows to implement infinitely many arbitrary functions of single argument into solution of the nonlinear PDE satisfying this criterion. Our approach to theproblem ofconstruction of theparticular solutions to the Self-dualtype S-integrable PDEs is most similar to the method of hydrodynamic reductions. However, if it is applicable, we are able to describe the bigger solutions space. Namely, number of arguments in the arbitrary functions implemented into solution has no restriction and is defined by the particular nonlinear PDE. But we have no criterion for applicability of this algorithm to the given nonlinear PDE, unlike the method of hydrodynamic reductions. We exhibit a class of solutions with wave profile breaking to the S-integrable PDEs selected above. This phenomenon is not surprising, since, simultaneously, these solutions are solutions of the well established class of first order quasilinear PDEs in lower dimension, which are known to possess solutions spaces implicitly described by systems of non-differential equations [29, 30]. Remark that solutions with breaking of wave profile to dispersion-less Kadomtsev-Pertviashvili equation (dKP) (which is compatibility condition of appropriate pair of vector fields) have been obtained and described in [31] using ISTM. In the rest of Introduction we, first of all, recall the method of characteristics applicable to a certain type of matrix first order quasilinear PDEs, Sec.1.1. Then we remind some important properties of theSelf-dual type nonlinear S-integrable PDEs, Sec.1.2. Basic novalties appearing in this paper are underlined in Sec.1.3. Finally, we represent the brief contents of the paper in Sec.1.4. 1.1 First order quasilinear PDEs integrable by the method of char- acteristics As it was mentioned above, we associate the given Self-dual type S-integrable nonlinear PDE with the family of lower dimensional first order matrix PDEs integrable by the method of characteristics. Let us describe this family of PDEs. Throughout this paper we will write superscripts in parentheses in order to distinguish them from powers. It is well known that the scalar first order PDE N u + u ρ(k)(u) = ρ(u), u : Rn+1 R, (1) t x k → k=1 X can be solved by the method of characteristics [18]. For instance, solution u(x,t) of (1) for 2 ρ = 0 is defined implicitly by the non-differential equation u = f x ρ(1)(u)t,..,x ρ(N)(u)t . (2) 1 N − − Vector generalizations of equation(cid:0) (1), i.e. the systems of seve(cid:1)ral coupled first order scalar PDEs, have been investigated in set of papers [19, 21, 32, 33] using generalized hodograph method in most cases. Recently [29] the matrix generalization of eq.(1) has been solved using algebraic approach: N w + w ρ(k)(w) = ρ(w)+[w,Tρ˜(w)], (3) t x k k=1 X wherew istheunknownQ QmatrixfunctionoftheN+1independent variables(x ,..,x ,t) 1 N RN+1, T is any constant d×iagonal matrix, [ , ] is the usual commutator between matrices, an∈d · · ρ(k), ρ, ρ˜ : R R, k = 1,..,N are N + 2 arbitrary scalar functions representable by the → positive power series: ρ(j)(z) = α(j;i)zi, ρ(z) = α(i)zi, ρ˜(z) = α˜(i)zi, (4) i≥0 i≥0 i≥0 X X X where α(j;i), α(i) and α˜(i) are scalar constants, so that the quantities ρ(k)(w), k = 1,..,N, ρ(w) and ρ˜(w) are well-defined functions of the matrix w. Remark. We may always apply the following change of variable t and field w N ∂ + α(j;0)∂ ∂ , w e−α˜(0)Ttweα˜(0)Tt (5) t xj → t → j=1 X which is equivalent to putting α(j;0) = 0, α˜(0) = 0 (6) in the eqs.(4) without loss of generality. Solutionsspace to thematrixequation (3) withρ = 0is implicitly described by thefollowing algebraic equation: w = (7) αβ Q e−Tαρ˜(w)tF x I ρ(1)(w)t,...,x I ρ(N)(w)t eTγρ˜(w)t , αγ 1 N − − γβ Xγ=1(cid:16) (cid:16) (cid:17) (cid:17) α = 1,...,Q, β = 1,...,Q, I is Q Q identity matrix. × If ρ = 0, then eq.(3) has infinitely many commuting flows. In this case, it is convenient to replace eq.(3) by the next equation N w + w ρ(mk)(w) = [w,Tmρ˜(m)(w)], m = 1,2,..., (8) tm xk k=1 X where index m enumerates commuting flows, ρ(mk) are arbitrary scalar functions representable by a positive power series ρ(mj)(z) = α(mj;i)zi, ρ˜(m)(z) = α˜(m;i)zi, (9) i≥0 i≥0 X X 3 α(mj;i), α˜(m;i) are scalar constants, T(m) are diagonal matrices. Note, that the transposed equation N w˜ + ρ(mk)(w˜)w˜ = [ρ˜(m)(w˜)Tm,w˜], w˜ = wT, m = 1,2,... (10) tm xk k=1 X is also integrable. Equation (3) as well as its generalizations whose solutions spaces may be completely de- scribed implicitly by the system of non-differential equations will be referred to as PDE0 in this paper for the sake of brevity. Using either the algebraic manipulations (Sec.2) or the modified version of the dressing method (Sec.3), we will show that PDE0s (8) may be viewed as lower dimensional reductions of the classical Self-dual type S-integrable PDEs. These PDEs are referred to as PDE1s in this paper. 1.2 Family of PDE1s PDE1s may be viewed as compatibility conditions of the following overdetermined system of linear PDEs for some spectral function V(λ;x) (spectral system): N N˜ Nm(ma)x V (λ;x)+ λnmiV (λ;x)+ λn˜miV(λ;x)T(i) = λiV(λ;x)q(mi)(x), (11) tm xi i=1 i=1 i=0 X X X m = 1,2. Here all functions are Q Q matrix functions, λ is a complex spectral parameter, n and mi n˜ are some integers, N×(m) = max(n ,n˜ , i = 1,...,N, i = 1,...,N˜); q(mi)(x) are mi max mi1 mi2 1 2 some matrix functions of x = (x ,x ...,t ,t ...). Throughout this paper we use Greek letters in 1 2 1 2 the lists of arguments for spectral parameters. The feature of this system is that its LHS is a first order differential operator having coefficients which are polynomial in λ and independent on the vector parameter x, i.e. x-dependent potentials appear only in the RHS, which is also polynomial in λ. Compatibility condition of any pair of eqs.(11) results in PDE1 for the functions q(mi). Note that eq.(11) uses only multiplication by the spectral parameter λ and does not involve derivatives of the spectral function with respect to the spectral parameter. For this reason such S-integrable PDEs as dispersion-less Kadomtsev-Petviashvili equation (dKP) and Heavenly equation may not be considered in the framework of the represented dressing algorithm. Remark, that the spectral system (11) may be replaced by the equivalent one N N˜ ∂ V(λ;x)+ Anmi(λ,ν) ∂ V(ν;x)+ An˜mi(λ,ν) V(ν;x)T(i) = (12) tm ∗ xi ∗ i=1 i=1 X X Nm(ma)x Ai(λ,ν) V(ν;x)q(mi)(x), ∗ i=0 X where function A(λ,µ) replaces multiplication by λ and we define operator Ai as follows: Ai = A A. (13) ∗···∗ i | {z } 4 Emphasise that A is independent on x. The eq.(12) reduces to the eq.(11) if A(λ,ν) = λδ(λ − µ)I, where I is Q Q identity matrix. × Example 1: N-wave system. Linear spectral system reads in the simplest case: ∂ V(λ;x) A(λ,ν) V(ν;x)T(m) = V(λ;x)[T(m),w(0)(x)], m = 1,2. (14) tm − ∗ Compatibility condition of this system yields the next PDE1: [T(1),w(0)] [T(2),w(0)]+[[T(2),w(0)],[T(1),w(0)]] = 0. (15) t2 − t1 (0) (0) Most physical applications require reduction w = w¯ , where α = β and ”bar” means αβ βα 6 complex conjugate. This equation is well known as (1+1)-dimensional N-wave equation [10]. We consider eq.(15) without reductions. Example 2: Pohlmeyer equation. Another example is assotiated with the following spec- tral system: ∂ V(λ;x)+A(λ,ν) ∂ V(ν;x) = V(λ;x)w(0)(x), m = 1,2. (16) tm ∗ xm xm Compatibility condition of this system yields the next PDE1: w(0) w(0) = [w(0),w(0)], (17) x1t2 − x2t1 x1 x2 which may be written in a different form: (J−1J ) (J−1J ) = 0, (18) t1 x2 − t2 x1 w(0) = J−1J , w(0) = J−1J . x1 t1 x2 t2 This is Pohlmeyer equation [34, 35, 36]. Most applications in Physics is assotiated with its reduction J+ = J, detJ = 1, which yields (Anty)Selfdual Yang-Mills equation (ASDYM). ± This equation has been studied by many authors: [37, 38, 39, 40, 41, 42, 43, 44] (instanton solutions), [45] (merons), [46, 47] (finite gap solutions). (2+1)-dimensional version of ASDYM assotiated with the reduction ∂ J = ∂ J, ∂ w(0) = ∂ w(0) (19) x2 t1 x2 t1 is also well known: w(0) w(0) = [w(0),w(0)] or (20) t1t1 − t2x1 t1 x1 (J−1J ) (J−1J ) = 0. t1 t1 − t2 x1 It has been studied in [48, 49] (Initial Value Problem), [50, 51] (localized solutions). Here we will deal with eqs.(17) and (20a) without any reduction for the field w(0). 5 1.3 Basic novalties There are two manifolds of particular solutions to PDE1s. First manifold is assotiated with the uniquely solvable linear integral equation for some spectral function (Zakharov-Shabat method ¯ [52], classical ∂-problem [53]). We concentrate on the second manifold of solutions which is described implicitly by a system of non-differential equations (see Sec.2). Simultaneously, this manifold is solution manifold to appropriate lower dimensional PDE0. Proper choice of initial data leads to wave profile breaking. Such solutions for Pohlmeyer equation will be discussed. In turn, second manifold of solutions has a sub-manifold of explicit solutions. Regarding (1+1)- dimensional N-wave equation, only such solutions from the second manifold are available. It is likely that the same conclusion is valid for any other (1+1)-dimensional S-integrable model. Itisremarkable, thatthereisaversion ofthedressing method, whichjoinsbothmanifoldsof solutions to DPE1s, Sec.3. This fact is remarkable, because the second manifold is beyond the scope ofthe classical dressing method. Inaddition, this fact demonstrates that two significantly different classes of PDEs (S-integrable PDE1s and integrable by the method of characteristics PDE0s) arecombined by thedressing method. There areexamples ofsimilar combinations. For instance, dressing method combining S- and C-integrable models has been suggested in [54]; dressing method combining C-integrable equations and equations integrable by the method of characteristics has been proposed in [30]. All these examples tell us about flexibility of the dressing method which is important for development of the uniform method for integration of multidimensional nonlinear PDEs. Usually, we will use notations PDE1(t ,t ;w(0)), PDE0(t ;n ,w) and PDE0(t ;k ,w˜) m1 m2 m 0 m 0 with arguments reflecting field as well as derivative(s) of field with respect to variable(s) t m appearing in PDE. Of course, PDE0s and PDE1s involve also derivatives with respect to x . i However, we do not represent them in the lists of arguments. Here w(x) is n Q n Q matrix 0 0 × and w˜(x) is k Q k Q matrix block functions: 0 0 × w(0)(x) w(1)(x) w(n0−2)(x) w(n0−1)(x) ··· I 0 0 r(1) ··· w(x) = 0 I 0 r(2) , (21) ··· .. .. .. .. .. . . . . . 0 0 I r(n0−1) ··· w˜(0)(x) I 0 0 − ··· w˜(1)(x) 0 I 0 − ··· . . . . . w˜(x) = .. .. .. .. .. , w˜(0) = w(0), w˜(k0−2)(x) 0 0 I − ··· w˜(k0−1)(x) r˜(1) r˜(2) r˜(k0−1) − ··· where I and 0 are Q Q identity and zero matrices, w(i) and w˜(i) are Q Q matrix functions, × × n and k are arbitrary integers, r(i), r˜(i), i = 1,...,n 1 are arbitrary constant Q Q 0 0 0 − × matrices (if T(i) = 0 in eqs.(25,26)) or arbitrary diagonal Q Q matrices (if T(i) = 0). Each × 6 PDE1(t ,t ;w(0)) written for the matrix field w(0) and involving derivatives with respect to t 1 2 1 and t possesses a family of PDE0(t ;n ,w) and PDE0(t ;k ,w˜), m = 1,2, n ,k = 1,2,..., 2 m 0 m 0 0 0 as lower dimensional reductions. Of course, PDE0(t ;n ,w) is compatible with PDE0(t ;n ,w), 1 0 2 0 as well as PDE0(t ;n ,w˜) is compatible with PDE0(t ;n ,w˜). 1 0 2 0 For instance, in Sec.2 we will derive the following PDE0s assotiated with N-wave equation 6 (Sec.2.1.1): w [w,T(m)w] = 0, (22) tm − n0 w˜ [w˜T˜(m),w˜] = 0, m = 1,2, tm − k0 T(m) = diag(T(m) T(m)), T˜(m) = T(m) n ··· n − n n and PDE0s assotiated with Pohlmeyer equatio|n (Se{cz.2.1.2}): w +w w = 0, (23) tm xm w˜ +w˜w˜ = 0, m = 1,2. tm xm If reduction (19) is imposed, then eq.(23) reduces to w +w w = 0, w +w w = 0, (24) t1 x1 t2 t1 w˜ +w˜w˜ = 0, w˜ +w˜w˜ = 0. t1 x1 t2 t1 General forms of PDE0(t ;n ,w) and PDE0(t ;k ,w˜), m = 1,2, which are lower dimensional m 0 m 0 reductions of some PDE1(t ,t ;w(0)) are following: 1 2 N N˜ w + w ρ(mi)(w) = [w,T(i)ρ˜(mi)(w)], m = 1,2, (25) tm xi n0 i=1 i=1 X X N N˜ w˜ + ρ(mi)(w˜)w˜ = [ρ˜(mi)(w˜)T˜(i),w˜], m = 1,2, (26) tm xi k0 i=1 i=1 X X where functions ρ˜(mi) are defined by the series similar to the eq.(9): ρ˜(mj)(z) = α˜(mj;i)zi, (27) i≥0 X and α˜(mj;i) are scalar constants. Remarks on the origin of eqs.(25) and (26) will be given in Sec.3.3, see n.3,5 and n.4,6 respectively. Eq.(26) has the form of transposed equation (25) with replacements w˜ = wT, T(m) T˜(m). (28) n0 → k0 Because of this similarity, we will deal with PDE1(t ,t ;w(0)) and PDE0(t ;n ,w), m = 1,2. 1 2 m 0 Regarding PDE0(t ;k ,w˜), only some details will be given. m 0 Remark 1. Although any PDE1 is compatibility condition of eqs.(12), we are not able to represent general form of PDE1s explicitly, unlike general form of PDE0, eqs.(25,26). Remark 2. The eq.(25) defers from the eq.(8) by the sum in the RHS. However, eq.(25) may be treated by the methods developed in [29, 30] for eq.(8). Remark 3. The eqs.(22,23,25,26) (where m = 1,2) have infinitely many commuting flows corresponding to m > 2 in these equations. 7 1.4 Brief contents In the next section (Sec.2) we give algebraic description of PDE1(t ,t ;w(0)), PDE0(t ;n ,w) 1 2 m 0 and PDE0(t ;k ,w˜). Derivation of PDE1 (namely, (1+1)-dimensional N-wave and (3+1)- and m 0 (2+1)- dimensional Pohlmeyer equations) assotiated with appropriate PDE0(t ;n ,w) and m 0 PDE0(t ;k ,w˜), m = 1,2 is given in Sec.2.1. Assotiated linear overdetermined systems have m 0 been described therein as well. Using slightly modified results of [29, 30] we represent the non- differential matrix equations implicitly describing the solutions spaces to the above PDE0s and assotiated solutions manifolds to the appropriate PDE1s, Sec.2.2. We derive some manifolds of explicit solutions to N-wave and Pohlmeyer equations and describe solutions with wave profile breaking for (2+1)- and (3+1)-dimensional Pohlmeyer equations. A new version of the dressing method describing PDE0s and their relations with PDE1s is proposedinSec.3. WehavederivedexamplesofPDE0sandexamplesofPDE1s(followingSec.2, N-wave and Pohlmeyer equations are taken as examples of PDE1s) together with appropriate linear overdetermined systems (spectral systems), Sec.3.1. Classical solutions manifolds to PDE1s (Sec.3.2.1) and solutions manifolds with wave profile breaking to PDE1s and PDE0s (Sec.3.2.2) have been derived in Sec.3.2. Although this version of the dressing method is mostly straightforward in spirit of the derivation of PDE1s, it does not supply full solutions spaces to PDE0s, unlike the method of characteristics. The second version of the dressing algorithm exhibits fullness of the solutions spaces to PDE0s, Sec.3.3. Conclusions are given in Sec.4. 2 Algebraic description of PDE1s and PDE0s We consider three examples of PDE1(t ,t ;w(0)) (namely, eqs.(15) and (17) together with its 1 2 (2+1)-dimensionalreduction(20a))andappropriatefamiliesofPDE0(t ;n ,w)andPDE0(t ;k ,w˜), m 0 m 0 m = 1,2, in the framework of the algebraic approach, see Secs.2.1.1,2.1.2. The solutions spaces to PDE1(t ,t ;w(0)) and PDE0(t ;n ,w), m = 1,2, will be investigated in Sec.2.2. 1 2 m 0 2.1 Derivation of PDE1s and appropriate families of PDE0s 2.1.1 (1+1)-dimensional N-wave equation It iswell known, that any S-integrableequation can berepresented assystem of two commuting discrete chains. Mostly explicitly this representation is given in Sato approach to integrability, see for instance, [55]. However, any known dressing method gives rise to this representation. We show, that N-wave equation can be derived from the following pair of commuting discrete chains with two discrete parameters: w(kn) = w(k(n+1))T(m) T(m)w((k+1)n) +w(k0)T(m)w(0n), m = 1,2, (29) tm − supplemented by the next non-differential relation among w(ij): w((k+1)n) = w(k(n+1)) +w(k0)w(0n) (30) (see also Sec.3.1.3, eqs.(150,151) with s(m) = 0). In fact, first of all remark, that eq.(29) gives rise to two alternative discrete chains with single discrete parameter in view of (30). The first 8 chain follows after putting k = 0 and eliminating w(1n) using eq.(30): w(n) [w(n+1),T(m)]+[T(m),w(0)]w(n) = 0, m = 1,2, (31) tm − where w(n) = w(0n), n = 0,1,.... (32) The second discrete chain follows after putting n = 0 and eliminating w(k1) from (29) using (30): w˜(k) [w˜(k+1),T(m)]+w˜(k)[w˜(0),T(m)] = 0, m = 1,2, (33) tm − where w˜(k) = w(k0), k = 0,1,..., w˜(0) = w(0). (34) Finally, in order to write PDE1 for the function w(0), we fix n = 0 in (31) and eliminate w(1), or fix k = 0 in (33) and eliminate w˜(1). In both cases we will end up with the same equation (15). Along with PDE1(t ,t ;w(0)) (15) we may derive the family of PDE0(t ;n ,w), m = 1,2, 1 2 m 0 from the eqs.(31) imposing the reduction n0−1 w(n0)(x) = w(i)(x)r(i), (35) i=0 X where n is an arbitrary integer and r(i) are arbitrary diagonal constant matrices. Similarly, we 0 may derive PDE0(t ;k ,w˜), m = 1,2, from the eq.(33) with the reduction m 0 k0−1 w˜(k0)(x) = r˜(i)w˜(i)(x), w˜(0) = w(0), (36) i=0 X where k is arbitrary integer and r˜(i) are arbitrary diagonal constant matrices. Both (35) 0 and (36) are closures of the chains (31) and (33) respectively. However, parameters r(0) and r˜(0) may be removed from the PDE0s by the shifts of fields w(n0−1) + r(0) w(n0−1) and → w˜(k0−1) +r˜(0) w˜(k0−1). Thus hereafter we take − → − r(0) = r˜(0) = 0 (37) without loss of generality. Parameters n , k , r(i) and r˜(i) appear in the definitions of w and w˜, 0 0 eqs.(21). The simplest explicit examples of PDE0(t ;n ,w) and PDE0(t ;k ,w˜) are following: m 0 m 0 PDE0(t ;1,w) : w(0) +[T(m),w(0)]w(0) = 0, m = 1,2, (38) m tm PDE0(t ;2,w), r(1) = 0 : w(0) [w(1),T(m)]+[T(m),w(0)]w(0) = 0, (39) m tm − w(1) +[T(m),w(0)]w(1) = 0, m = 1,2, tm PDE0(t ;1,w˜) : w˜(0) +w˜(0)[w˜(0),T(m)] = 0, (40) m tm PDE0(t ;2,w˜), r˜(1) = 0 : w˜(0) [w˜(1),T(m)]+w˜(0)[w˜(0),T(m)] = 0, m tm − w˜(1) +w˜(1)[w˜(0),T(m)] = 0, m = 1,2. tm 9 It is not difficult to observe that the above PDE0(t ;n ,w) and PDE0(t ;k ,w˜) with arbitrary m 0 m 0 n and k , may be written in the form (22a) and (22b) respectively, where w and w˜ are defined 0 0 by the formulae (21). Since the eq. (22b) follows from the eq.(22a) after transposition with replacements (28), it is enough to study one of them, say, eq.(22a). Slightly modifying result of [30], we describe the solutions space to the eq.(22a) with arbitrary integer n by the next implicit algebraic equation: 0 wαβ = n0Q e−m2=1(Tn(m0))αwtmFαγ(w) em2=1(Tn(m0))γwtm , (41) P P γβ Xγ=1(cid:16) (cid:17) α,β = 1,...,n Q, 0 where F(z ) is n Q n Q matrix function of single argument. Details of derivation of this 0 0 0 × formula by the dressing method are given in Secs.3.2.2 and 3.3. However, the structure of w tells us that F must have the following structure: arbitrary scalar function of arguments, α Q F (z ) = ≤ , (42) αβ 0 δ z , α > Q αβ 0 (cid:26) sothatequation(41)becomesanidentityforα > Q. Thus, thesquarematrixalgebraicequation (41) reduces to the rectangular matrix equation with α = 1,...,Q and β = 1,...,n Q. The 0 scalar functions of single argument F (z ), α = 1,...,Q, β = 1,...,n Q are arbitrary. The αβ 0 0 simplest case n = 1 yields w w(0), T(m) T(m): 0 ≡ 1 ≡ Q 2 2 wα(0β) = e−mP=1Tα(m)w(0)tmFαγ(w(0)) emP=1Tγ(m)w(0)tm γβ, α,β = 1,...,Q. (43) Xγ=1(cid:16) (cid:17) Assotiated linear overdetermined system of PDEs. As it was mentioned above, the linear spectral problem for N-wave equation is given by the eqs.(14), where V(λ;x) is, generally speaking, a rectangular lQ Q matrix function, l is some integer (to anticipate, l is assotiated × with the dimension of the kernel of the integral operator in eq.(114); l = 2 in Secs.3.1,3.2). One can extend this spectral system introducing discrete parameter n by the following formulae: A(λ,ν) V(n)(ν;x) = V(n+1)(λ;x)+V(0)(λ;x)w(n)(x), (44) ∗ V(n)(λ;x) A(λ,ν) V(n)(ν;x)T(m) = V(0)(λ;x)[T(m),w(n)(x)], (45) tm − ∗ n = 0,1,2,...,m = 1,2. which becomes eq.(14) if n = 0 and V V(0). This extension can be derived formally, for ≡ instance, using a version of the dressing method developed in Sec.3, see Sec.3.1.3 eq.(149) with s(m) = 0. Let us consider the reduction (35,37) for the fields w(n), which causes the appropriate re- duction for the spectral functions V(n): n0−1 V(n0) = V(i)r(i). (46) i=1 X In view of this reduction, we may introduce the block-vector spectral function V = [V(0) V(n0−1)]. (47) ··· 10